CN101494636B - Method and apparatus for ordering data based on rapid IO interconnection technology - Google Patents

Abstract

The invention discloses a data sorting method based on fast IO interconnection technology. Three groups of first-in-first-out queues are preset for storing response packets with different priorities. The method includes: sending requests in units of request packet groups package; receive the response packet group corresponding to the request packet group from the destination, and store the response packets of the response packet group into three groups of first-in-first-out queues in sequence according to the priority of the response packets and the order of arrival at the source In: after receiving the response packet groups, read and send the response packets from the three groups of first-in-first-out queues sequentially according to the order of the transmission identifiers. The present invention effectively corrects the response packet transmission sequence of the Rapid IO interface on the premise of using less buffer resources and consuming less time delay, solves the problem of response packet disorder in the existing high-speed data processing system, and improves high-speed data processing. Processing system performance.

Description

A data sorting method and device based on fast IO interconnection technology

technical field

The invention relates to the interconnection technology of embedded systems, in particular to a data sorting method and device based on fast IO interconnection technology.

Background technique

Rapid IO (Rapid IO Architecture) is a set of open high-bandwidth full-duplex cascading solutions for chip-level and board-level interconnection developed by the Rapid IO Trade Association (Rapid IO Industry Association) in December 2001. Its performance can reach 10Gb/s or higher. It is a low-latency, memory-address-based protocol that is scalable, reliable, multiprocessing capable, and transparent to application software.

The packet types of the Rapid IO protocol can be divided into maintenance packet read and write, read request, read response, and write operations. Rapid IO read operation mode: the source initiates a request, and the destination responds, that is, the read request is first sent from the source to the destination in the form of packet exchange, and the destination sends the response packet to the source, thus realizing a complete operation.

In the current mainstream CPU with Rapid IO interface, there are usually 4 sets of Buffers (buffers) at the sending end of the Rapid IO interface, corresponding to sending data packets with 4 priority levels. The deadlock prevention rules of the Rapid IO protocol stipulate that "the priority of the packet carrying the response should be at least one level higher than that of the corresponding request packet", so the priority of the response packet can only have three levels: 1, 2, and 3.

The response mechanism of the destination CPU to the request packet is: according to the address and data length requested by the request packet, the corresponding data is packaged to generate a corresponding response packet. Due to the small number of bytes in the request packet, the packet transmission speed is higher than that of the corresponding response packet. Therefore, during the packaging process of the response packet, if the low-priority Buffer is filled, the newly generated response packet will be stored in the higher-priority buffer. In the buffer of the first level, this will cause the priority of the response packet in the later order to be higher than the response packet in the earlier order. The transaction and packet transmission ordering rules in the Rapid IO protocol stipulate that "the physical layer of the endpoint processing component port should ensure that the high-priority request transaction received from the physical layer of the processing component is forwarded before the low-priority request transaction, and the low-priority request transaction Priority packets cannot outnumber higher-priority packets". In this way, the following problem will occur during the transmission of the response packet: the CPU will first send the data packets with the lower order but with the higher priority, and then send the data packets with the earlier order but with the lower priority.

At present, the solution to this problem is usually to reduce the data packaging speed and slow down the sending rate by the destination CPU. However, in a high-speed data processing system sensitive to data order, the reduction in sending rate will lead to a decrease in system performance, and the rate is too fast. There will be out-of-sequence phenomenon, which will lead to system errors, which will seriously affect the application of Rapid IO interface in high-speed systems.

Contents of the invention

In view of the above analysis, the present invention aims to provide a data sorting method and device based on fast IO interconnection technology to solve the problem of out-of-order response packets in high-speed data processing systems in the prior art.

The purpose of the present invention is achieved through the following technical solutions.

The present invention provides a data sorting method based on fast IO interconnection technology. Three sets of first-in-first-out queues are preset for storing response packets with different priorities. The method includes:

Step A: sending request packets in units of request packet groups;

Step B: Receive the response packet group corresponding to the request packet group from the destination, and store the response packets of the response packet group into three first-in-first-out groups in sequence according to the priority of the response packets and the order of arrival at the source in the queue;

Step C: After receiving the response packet groups, read and send the response packets from the three groups of first-in-first-out queues sequentially according to the order of the transmission identifiers.

Further, the step A specifically includes:

According to the requirements of the system, the request package is divided into groups;

First send a part of the request packets in the request packet group;

After receiving the first response packet fed back by the destination, send the remaining request packets.

Further, when two groups of buffer units are preset, and each group of buffer units includes three groups of first-in-first-out queues, the step B specifically includes:

Carry out parity counting to the response packet group corresponding to the request packet group received from the destination end, and adopt the mode of ping-pong operation to select the buffer unit for storing the response packet group;

According to the priority of the response packets and the order in which they arrive at the source, the response packets of the response packet group are sequentially stored in three groups of first-in-first-out queues, and at the same time, another group of buffer units is cleared.

The length of the first-in-first-out queue is equal to the number of request packet groups.

The present invention also provides a data sorting device based on fast IO interconnection technology, including:

The sending module is used to divide the request packets into groups according to the requirements of the system, and send the request packets in units of request packet groups;

A sorting module, configured to receive the response packet groups corresponding to the request packet groups, and store the response packets of the response packet groups into three groups of first-in-first-out queues sequentially according to the priority of the response packets and the order of arrival at the source middle;

The reading module is configured to sequentially read and send response packets from the three groups of first-in-first-out queues according to the order of the transmission identifiers.

Further, the sorting module specifically includes:

a parity counter, configured to perform parity counting on the response packet group corresponding to the request packet group received from the destination;

The first selector is used to select the first buffer unit or the second buffer unit to store the response packet group according to the parity count;

The first buffer unit is configured to cache the response packet group in turn with the second buffer unit according to the selection result of the first selector;

The second buffer unit is configured to cache the response packet group in turn with the first buffer unit according to the selection result of the first selector;

The second selector is used for sending the response packet group obtained from the first buffer unit or the second buffer unit to the lower-level data processing unit.

In summary, the present invention provides a data sorting method and device based on fast IO interconnection technology, which can effectively correct the response packet transmission of the Rapid IO interface on the premise of using less buffer resources and consuming less time delay order, which solves the problem of out-of-order response packets in the existing high-speed data processing system, and improves the performance of the high-speed data processing system.

Description of drawings

Fig. 1 is a schematic flow chart of the method described in the embodiment of the present invention;

Fig. 2 is a schematic structural diagram of the sorting module in the embodiment of the present invention.

Detailed ways

Preferred embodiments of the present invention will be specifically described below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of the application and are used together with the embodiments of the present invention to explain the principles of the present invention.

Firstly, the method described in the embodiment of the present invention will be described in detail with reference to FIG. 2 .

Step 100: Open up three groups of FIFO (First Input First Output) spaces at the source in advance for storing response packets of three different priorities, and the depth of each FIFO is M.

Step 101, the source end sends request packets in units of groups; specifically, the source end divides the request packets into multiple request packet groups in groups according to system requirements, and each group includes M request packets. When sending a request packet to the destination, a group of request packets are sent per unit of time. The sending process is: send M-N packets first, and send the remaining N request packets after receiving the first response packet of the group.

Step 102: After receiving the request packet, the destination end sends a corresponding response packet.

Step 103: After the source end receives the response packets corresponding to the request packets it sent, it stores the response packets in different FIFO spaces sequentially according to the difference in receiving priority and the order of arrival.

Step 104: When the last response packet of the group is received at the source end, the transaction ID (transmission identification) value of the first packet of the FIFO with a priority of 1 is used as the reference, and the three FIFOs are selected according to the order of the Transaction ID The response packet is read in and sent to the next-level data processing unit for corresponding processing.

During the above processing, in order to improve efficiency, the embodiment of the present invention adopts a ping-pong operation. Specifically, two sets of buffer units are opened at the source in advance, and three sets of FIFO spaces are opened in each set of buffer units to store response packets with different priorities, and the depth of each FIFO is M.

At the source end, the parity count is performed on the received response packet group to determine the buffer unit used in the ping-pong operation. For example, when the first group of response packets is received, it is recorded as an odd number and stored in the first buffer unit; when the second group of response packets is received, it is recorded as an even number and stored in the second buffer unit unit; when receiving the third group of response packets, record it as an odd number and store it in the first buffer unit; and so on. Then, according to the priority of the response packets and the order in which they arrive at the source, the response packets of the response packet group are stored in three groups of first-in-first-out queues, and at the same time, another group of buffer units is emptied, and steps 103 and 104 are repeated. . The invention realizes the seamless buffering and processing of the response packet by adopting the ping-pong operation mode.

Next, the device described in the embodiment of the present invention is described in detail.

The device described in the embodiment of the present invention specifically includes:

The sending module divides the request packet into groups according to the requirements of the system, and sends the request packet to the destination end in units of request packet groups; the destination end packages the received request packet to generate a corresponding response packet, and Send the generated response packets in turn.

The sorting module receives the response packets corresponding to the request packets, and according to the priority of the response packets and the order of arrival at the source, sequentially stores the response packets of the response packet groups into three groups of first-in-first-out queues;

The reading module, after receiving a group of response packets, reads and sends the response packets sequentially from the three groups of first-in-first-out queues according to the order of the transmission identifiers of the response packets.

Here, in order to improve efficiency, the embodiment of the present invention adopts a ping-pong operation. Specifically, two groups of buffer units are opened at the source in advance, and three groups of FIFO spaces are opened in each group of buffer units to store response packets with different priorities. The depth of each FIFO is M, then The structure of the sorting module is as shown in Figure 2, specifically may include:

The parity counter is used to count the parity of the response packet group corresponding to the request packet group received from the destination end; each time a group of response packets is received, the parity counter counts once. For example, when the first group of response packets is received, it is recorded as an odd number and stored in the first buffer unit; when the second group of response packets is received, it is recorded as an even number and stored in the second buffer unit unit; when receiving the third group of response packets, record it as an odd number and store it in the first buffer unit; and so on.

The first selector is used to select the first buffer unit or the second buffer unit to store the response packet group according to the parity count;

The first buffer unit is configured to cooperate with the second buffer unit according to the selection result of the first selector to cache the response packet groups in turn;

The second buffer unit is configured to cooperate with the first buffer unit according to the selection result of the first selector to cache the response packet group in turn;

The second selector is used for sending the response packet group obtained from the first buffer unit or the second buffer unit to the lower-level data processing unit.

The specific process of the ping-pong operation can be: the response packet isochronously distributed to two groups of buffer units by the first selector, and when the parity counter is an odd number, the first group of response packets received is buffered into the first buffer unit unit; when the parity counter changes from odd to even, the second group of response packets received is buffered into the second buffer unit through the switching of the first selector, and the first group buffered by the first buffer unit is simultaneously The response packet is sent to the lower-level data processing unit for related processing through the selection of the second selector; when the parity counter changes from even to odd, the third group of response packets received is cached by switching the first selector again. The first buffer unit simultaneously sends the second group of response packets buffered by the second selector to the lower-level data processing unit for related processing through the switching of the second selector; and so on.

Of course, the present invention is not limited to one or two groups of buffer units. In order to achieve the purpose of fast sorting, several groups of buffer units can also be set up. The principle is similar to the ping-pong operation, and examples are not given here.

To sum up, the embodiment of the present invention provides a data sorting method and device based on the rapid IO interconnection technology, according to the Rapid IO specification "have the same source ID, the same destination ID, the same priority and ftype!=8( ftype indicates the type of the data packet) and the order of the packets is not changed during the transmission process", sort the response packets of each priority separately, and then send the sorted response packets to the lower-level processing unit according to the order of the Transaction ID. In the processing process of the embodiment of the present invention, in order to improve the efficiency, the seamless buffering and processing of data is completed by means of ping-pong operation.

The embodiment of the present invention effectively corrects the response packet transmission sequence of the Rapid IO interface on the premise of using less buffer resources and consuming less time delay, solves the problem of out-of-order response packets in the existing high-speed data processing system, and improves Performance of high-speed data processing systems.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present invention can easily think of changes or Replacement should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (6)

1. the data reordering method based on rapid IO interconnection technology is characterized in that, presets three groups of First Input First Outputs, is used for the respond packet of storage different priorities, and described method comprises:

Steps A: send the request bag take the request package as unit;

Step B: receive the response package corresponding with the described request package from destination, and according to the sequencing of priority and the arrival source of respond packet, the respond packet with described response package deposits in three groups of First Input First Outputs successively in;

Step C: after receiving described response package, according to the order of transmission sign, successively reading and sending respond packet from described three groups of First Input First Outputs.

2. method according to claim 1 is characterized in that, described steps A specifically comprises:

According to the demand of system, will ask bag to be divided take group as unit;

Send first the part request bag in the request package;

After the first respond packet that receives the destination feedback, remaining request bag is sent.

3. method according to claim 1 is characterized in that, when presetting two groups of buffer locations, when every group of buffer location comprised three groups of First Input First Outputs, described step B specifically comprised:

The response package corresponding with the described request package that receives from destination carried out parity count, and the buffer location that adopts the described response package of mode Selective storage of ping-pong operation to use;

According to the priority and the sequencing that arrives source of respond packet, the respond packet with described response package deposits in three groups of First Input First Outputs successively, empties simultaneously unit, other pool of buffer district.

4. the described method of any one in 3 according to claim 1 is characterized in that, the length of described First Input First Output equals to ask the number of package.

5. the data sorting device based on rapid IO interconnection technology is characterized in that, comprising:

Sending module is used for the demand according to system, will ask bag to be divided take group as unit, and sends the request bag take the request package as unit;

Order module be used for to receive the response package corresponding with the described request package, and according to the sequencing of priority and the arrival source of respond packet, the respond packet with described response package deposits in three groups of First Input First Outputs successively;

Read module is used for the order according to the transmission sign, successively reading and sending respond packet from described three groups of First Input First Outputs.

6. device according to claim 5 is characterized in that, described order module specifically comprises:

Odd-even counter is used for the response package corresponding with the described request package that receives from destination carried out parity count;

First selector is used for according to parity count, selects the first buffer location or the second buffer location to store described response package;

The first buffer location is used for the selection result according to first selector, with the second buffer location described response package of buffer memory in turn;

The second buffer location is used for the selection result according to first selector, with the first buffer location described response package of buffer memory in turn;

Second selector is used for and will sends to subordinate's data processing unit from the response package that the first buffer location or the second buffer location obtain.

Images